WO2012050705A1 - Égalisation automatique à l'aide d'un filtrage adaptatif par domaine de fréquence et d'une convolution rapide dynamique - Google Patents

Égalisation automatique à l'aide d'un filtrage adaptatif par domaine de fréquence et d'une convolution rapide dynamique Download PDF

Info

Publication number
WO2012050705A1
WO2012050705A1 PCT/US2011/051322 US2011051322W WO2012050705A1 WO 2012050705 A1 WO2012050705 A1 WO 2012050705A1 US 2011051322 W US2011051322 W US 2011051322W WO 2012050705 A1 WO2012050705 A1 WO 2012050705A1
Authority
WO
WIPO (PCT)
Prior art keywords
samples
block
filter
audio signal
frequency
Prior art date
Application number
PCT/US2011/051322
Other languages
English (en)
Inventor
Louis D. Fielder
David S. Mcgrath
Original Assignee
Dolby Laboratories Licensing Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dolby Laboratories Licensing Corporation filed Critical Dolby Laboratories Licensing Corporation
Priority to CN201180049284.8A priority Critical patent/CN103155591B/zh
Priority to EP11764382.5A priority patent/EP2628317B1/fr
Priority to US13/878,705 priority patent/US9084049B2/en
Publication of WO2012050705A1 publication Critical patent/WO2012050705A1/fr

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R3/00—Circuits for transducers, loudspeakers or microphones
    • H04R3/04—Circuits for transducers, loudspeakers or microphones for correcting frequency response
    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
    • H03H17/00—Networks using digital techniques
    • H03H17/02—Frequency selective networks
    • H03H17/06—Non-recursive filters
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
    • H04L25/03006—Arrangements for removing intersymbol interference
    • H04L25/03178—Arrangements involving sequence estimation techniques
    • H04L25/03248—Arrangements for operating in conjunction with other apparatus
    • H04L25/03286—Arrangements for operating in conjunction with other apparatus with channel-decoding circuitry

Definitions

  • the present invention pertains generally to audio signal filtering and pertains more specifically to techniques that may be used to adapt an equalization filter to have a desired frequency response.
  • Audio equalization filters are used in a variety of audio signal processing systems to modify an audio signal so that the transfer function of the system conforms to a desired frequency response.
  • an equalization filter may be used to compensate for frequency-response characteristics of electronic and acoustic components of an audio playback system so that the overall system transfer function is spectrally flat.
  • the frequency response of an equalization filter may be static or dynamic; however, dynamic or adaptive equalization (AEQ) filters are preferred for many applications because they can compensate for changing response characteristics of a system.
  • AEQ filters operate by minimizing a measure of difference between two time-domain signals such as system input and output signals, and they are responsive to both magnitude and phase differences between the signals.
  • AEQ filters whether static or dynamic, typically require an initial setup or calibration process to determine system response characteristics for both magnitude and phase so that the values of one or more parameters of the equalization filter can be set properly.
  • an initial setup process is typically required to determine a variety of characteristics very accurately such as equipment signal processing delays and acoustic signal propagation delays so that phase errors due to temporal misalignment can be minimized. If the initial setup process is not done properly, temporal alignment errors may cause an conventional AEQ filter to operate poorly and become unstable under certain conditions.
  • Fig. 1 is a schematic block diagram of one exemplary audio playback system that adapts the frequency response characteristics of an equalization filter.
  • Figs. 2 and 3 are schematic block diagrams of exemplary implementations of the analyzer in the system of Fig. 1.
  • Fig. 4 is a schematic block diagram of an exemplary implementation of the validity- measure generator in the system of Fig. 1.
  • Fig. 5 is a schematic block diagram of an exemplary implementation for the log average cross correlation calculator and the log average autocorrelation calculator in the analyzer of Fig. 3.
  • Figs. 6 and 7 are schematic block diagrams of exemplary implementations of the control- signal generator in the system of Fig. 1.
  • Fig. 8 is a schematic block diagram illustrating block operations for one
  • Fig. 9 is a schematic diagram of signal block structures.
  • Fig. 10 is a schematic diagram of processing steps for the static filter shown in Fig. 8.
  • Fig. 11 is a schematic block diagram of one implementation of the adaptive equalization filter in the system of Fig. 1.
  • Fig. 12 is a schematic diagram of processing steps for the adaptive filter shown in Fig. 11.
  • Fig. 13 is a schematic block diagram of an alternative implementation of the adaptive equalization filter in the system of Fig. 1.
  • Fig. 14 is a schematic block diagram of a device that may be used to implement various aspects of the present invention. MODES FOR CARRYING OUT THE INVENTION
  • Fig. 1 is a schematic block diagram of an audio playback system 1 that incorporates various aspects of the present invention. The diagram does not illustrate a complete system.
  • the illustration omits components like compact disc players and receivers that can provide audio signals for playback. These types of features are not needed to practice or to understand the present invention. A few alternative implementations are discussed following a general description of the system.
  • the audio playback system 1 includes an AEQ filter 100 that is applied to an audio signal 5, filters the audio signal 5 according to its current frequency response characteristics, and passes a filtered audio signal 195 to the driver 210.
  • the AEQ filter 100 may be implemented in a wide variety of ways including the use of one or more finite impulse response (FIR) and infinite impulse response (IIR) filters that are selected or adapted to have a frequency response that is equal to or approximates a specified frequency response.
  • FIR finite impulse response
  • IIR infinite impulse response
  • the filter may be applied to the entire bandwidth or only part of the bandwidth of the audio signal.
  • the driver 210 generates a signal in response to the filtered audio signal that is capable of driving an acoustic transducer such as a loudspeaker.
  • the driver 210 may be an audio power amplifier, for example, and may be implemented in any way that may be desired.
  • the driver 210 may be important in many implementations but it is not essential to practice the present invention.
  • the signal 215 generated by the driver 210 is passed to the acoustic output transducer 220.
  • the acoustic output transducer 220 generates the sound field 225 in response to the signal 215.
  • the acoustic output transducer 220 illustrated in the figure may be implemented by one or more distinct transducers, and these one or more transducers may be implemented by essentially any technology that may be desired. Essentially any type of loudspeaker or headphone transducer may be used but no particular type of output transducer is essential.
  • the acoustic input transducer 230 generates a detection audio signal 235 in response to the sound field 225.
  • the acoustic input transducer 230 illustrated in the figure may be implemented by one or more distinct transducers, and these one or more transducers may be implemented by essentially any technology that may be desired. For example, essentially any type of microphone may be used but no particular type of input transducer is essential.
  • the acoustic input transducer 230 should be located near the location where one or more listeners are expected to be located. If the acoustic output transducer 220 is incorporated into a headphone, the acoustic input transducer 230 should be located inside any cup or acoustic shield of the headphone near the ear of a listener.
  • the delay 250 receives a signal from some point along the signal processing path as illustrated and generates a delayed audio signal 255 that is a delayed replica of its input signal.
  • the delayed audio signal 255 is aligned in time with the corresponding detection audio signal 235.
  • the amount of delay needed to obtain proper alignment is establishing during an initial setup of the system.
  • the present invention is usually able to achieve very good results for larger errors in alignment than is possible using conventional methods.
  • the delay 250 may be implemented in any way that may be desired but it is anticipated that digital implementations will be preferred for many applications.
  • the amount of delay imposed by the delay 250 is set approximately equal to the total signal processing and propagation delay from the input to the AEQ filter 100 to the output of the acoustic input transducer 230. In many implementations, this total delay is approximately equal to a sum of the signal processing delay through the AEQ filter 100, the sound field 225 propagation delay from the acoustic output transducer 220 to the acoustic input transducer 230, and the processing delays for analog-to-digital conversion and buffering.
  • the control method used for this implementation is referred to below as the "feedback method" and allows the filter control system to adapt the frequency response characteristics of the AEQ filter 100 with that filter within a control loop.
  • the amount of delay imposed by the delay 250 is set approximately equal to a sum of the total signal processing and propagation delay from the input to the driver 210 to the output of the acoustic input transducer 230 and the processing delays for analog-to-digital conversion and buffering. In many implementations, this total delay is approximately equal to the sound field 225 propagation delay from the acoustic output transducer 220 to the acoustic input transducer 230.
  • the control method used for this implementation is referred to below as the "non- feedback method" and adapts the frequency response characteristics of the AEQ filter 100 with that filter outside the control loop. A similar situation exists if the delay 250 receives the signal 215 as its input.
  • the analyzer 300 receives the detection audio signal 235 and the delayed audio signal 255, obtains frequency-domain representations of the two signals, obtains a frequency- domain representation of a system target response 380 , and processes these three
  • a system transfer function 375 is derived from the frequency-domain representations of the detection audio signal 235 and the delayed audio signal 255, and the estimated spectral magnitude response correction signal 395 is generated by comparing this transfer function 375 to the system target response 380. Details of a few implementations are described below.
  • the control- signal generator 400 generates an equalization-filter control signal 495 in response to the estimated spectral magnitude response correction signal 395.
  • equalization-filter control signal 495 may be identical to or derived directly from the estimated spectral magnitude response correction signal 395; however, in preferred implementations, the equalization-filter control signal 495 is generated by applying a smoothing filter to a sequence of estimated spectral magnitude response correction signals 395 received from the analyzer 300.
  • the AEQ filter 100 adapts its frequency response characteristics in response to the equalization-filter control signal 495.
  • the validity-measure generator 500 is optional and may be used to improve AEQ filter adaptation in noisy environments.
  • the acoustic input transducer 270 should be located in a position where it can generate a second detection audio signal 275 in response to any ambient sounds that may be present.
  • the acoustic input transducer 270 may be implemented by one or more distinct transducers, and these transducers may be implemented by essentially any technology that may be desired. Essentially any type of microphone may be used. No particular type of input transducer is essential.
  • the second acoustic input transducer 270 should be located outside any cup or acoustic shield of the headphone away from the ear of a listener.
  • the validity-measure generator 500 compares signal levels or spectral characteristics of the second detection audio signal 275, the detection audio signal 235 and the delayed audio signal 255 and generates measures of validity 595 that are associated with components of the estimated spectral magnitude response correction signal 395.
  • control- signal generator 400 may modify magnitudes of one or more components of the estimated spectral magnitude response correction signal 395 in response to the associated measures of validity 595 so that adaptation of the AEQ filter 100 is less responsive to those components that are deemed to be less reliable.
  • the audio playback system 1 uses filterbanks to obtain frequency-domain representations of signals.
  • the analyzer 300 uses filterbanks to obtain frequency-domain representations of the detection audio signal 235 and the delayed audio signal 255.
  • These filterbanks may be implemented in essentially any way that may be desired that covers the frequency range of interest.
  • the filterbanks may be implementations of the Discrete Fourier Transform (DFT) and Inverse Discrete Fourier
  • IDFT Modified Discrete Fourier Transform
  • IMDFT Modified Discrete Fourier Transform
  • IMDFT Inverse Modified Discrete Fourier Transform
  • DCT Discrete Cosine Transform
  • IDCT Inverse Discrete Cosine Transform
  • QMF Quadrature Mirror Filter
  • CQMF Complex Quadrature Mirror Filter
  • FFT Fast Fourier Transform
  • the delay 250 receives its input signal from a signal path prior to the AEQ filter 100.
  • the amount of lag or delay imposed by the delay 250 is set so that corresponding intervals of the delayed audio signal and the detection audio signal are aligned in time when they are input to the analyzer 300.
  • the analyzer 300 processes a frequency-domain representation of a block or segment of the delayed audio signal 255, a frequency-domain representation of a block or segment of the detection audio signal 235, and a frequency-domain representation of a system target response 380 to generate an estimated spectral magnitude response correction signal 395.
  • This estimated correction signal 395 represents the change in frequency response of the AEQ filter 100 that is required for the audio playback system 1 to achieve an overall frequency response that matches the system target response 380.
  • One way that this may be done is by deriving a system transfer function 375 from a comparison of the spectral magnitudes in a block or segment of the delayed audio signal 255 with the spectral magnitudes in a corresponding block or segment of the detection audio signal 235, and then deriving an estimated correction signal 395 from a comparison of the system transfer function 375 to the system target response 380.
  • the filterbank 310a is implemented by an FFT that generates a block of complex- valued transform coefficients constituting a frequency-domain representation of the detection audio signal 235.
  • the log calculator 320a converts the complex values of these transform coefficients into logarithms of their magnitudes to facilitate subsequent
  • the operations of the filterbank apply an analysis window function to a sequence of overlapping blocks of signal samples and then apply a transform to the blocks of windowed samples to generate blocks of transform coefficients.
  • the FFT used in this implementation generates a block of transform coefficient pairs according to the following expression:
  • X(k) transform coefficient for spectral component k.
  • IFFT Inverse Fast Fourier Transform
  • the log calculator 320a generates a logarithmic
  • the filterbank 310b implemented by an FFT generates a block of complex-valued transform coefficients constituting a frequency-domain representation of the delayed audio signal 255.
  • the log calculator 320b converts the complex values of these transform coefficients into logarithms of their magnitudes.
  • the filterbank 310b and the log calculator 320b are implemented in the same way as the filterbank 310a and the log calculator 320a, respectively.
  • the subtractor 370 subtracts the logarithmic values of the transform coefficients received from the log calculator 320b from the logarithmic values of the transform
  • So(k) magnitude of transform coefficient k for the delayed audio signal
  • the system component represented by the system target response 380 in the figure provides a representation of the desired system frequency response or the system target response 380 in the log domain.
  • the subtractor 390 subtracts the log values of the system transfer function 375 from the log values of the system target response 380 to obtain an estimated spectral magnitude response correction signal 395.
  • the magnitude of the components of the estimated correction signal 395 are expressed in the log domain.
  • the subtractor 390 obtains a logarithmic representation of spectral magnitudes of the estimated spectral magnitude response correction signal 395 by calculations that may be expressed as:
  • TGT(k) audio playback system target response
  • T(k) system transform function in the linear domain
  • the estimated spectral magnitude response correction signal 395 is passed to the control- signal generator 400.
  • This correction signal is accurate over long time intervals but it can vary significantly from block to block due to variations in the audio signal 5, variations in ambient sounds, and variations in stochastic noise sources within the system itself.
  • the component values in the estimated correction signal 395 for any given block may differ significantly from the corrective values that should be provided to the AEQ filter 100.
  • a smoothing filter is applied to the estimated correction signal 395 to reduce or eliminate spurious variations that might otherwise cause adaptations of the AEQ filter 100 to generate audible artifacts. Smoothing is optional. In the exemplary implementations disclosed below, a smoothing filter is applied in the control- signal generator 400.
  • the control- signal generator 400 generates the equalization-filter control signal 495 that controls adaptation of the AEQ filter 100.
  • the control- signal generator 400 derives the equalization-filter control signal 495 by applying a smoothing filter to the estimated correction signal 395 to reduce system sensitivity to noise.
  • the control- signal generator 400 may also modify the values of selected
  • the system transfer function 375 and the estimated spectral magnitude response correction signal 395 are each represented by values arranged in a series of blocks. Abrupt changes in the values may occur from block to block. Temporal smoothing or low pass filtering of the components that constitute the equalization-filter control signal 495 can be used to eliminate abrupt changes in the frequency response of the AEQ filter 100 that might otherwise generate audible artifacts.
  • Fig. 6 illustrates one suitable implementation that includes a first-order smoothing function in the form of a leaky corrector. Empirical tests have shown that a first order filter is adequate for many applications but higher order smoothing filters may be used if desired.
  • the adaptation controller 410 generates a leak factor vector 415 and the vector multiplier 420 attenuates the magnitudes of the components in the estimated spectral magnitude response correction signal 395 according to the respective factors in the leak factor vector 415.
  • the vector adder 430 combines the attenuated components in the estimated spectral magnitude response correction signal 395 with corresponding components in the current desired AEQ filter response 450. This sum is delayed in the delay 440 by the block overlap interval, which is equal to Nil signal sample intervals, and then stored to become the current desired AEQ filter response 450.
  • the desired AEQ filter response 450 may be initialized at system startup by values that were established during system design or may be restored to the most recent desired response that was current when the system was shutdown.
  • the factors in the leak factor vector 415 may be adapted in response to signals that indicate the validity or reliability of the components in the estimated spectral magnitude response correction signal 395.
  • One or more of these components may not be a reliable indication of what response correction is needed if high-level ambient sounds or noise from any other source are present and detected by the acoustic input transducer 230.
  • the influence that a particular component of the estimated correction signal 395 has on the equalization-filter control signal 495 may be reduced or even eliminated if that component is deemed to be unreliable.
  • Preferred implementations assess the reliability of the components of the estimated correction signal 395 and use this assessment to control how the estimated correction signal 395 is used to adapt the AEQ filter 100.
  • One way that this may be done is to use the acoustic input transducer 270 and the validity-measure generator 500 as shown in Fig. 1 to generate the measures of validity 595 and to have the adaptation controller 410 modify, replace or attenuate the influence of respective components of the estimated correction signal 395 in response to these measures. This approach is described below.
  • the signal activity detector 520 analyzes the delayed audio signal 255 and generates a set of values 525 that indicate whether significant spectral components are present in the delayed audio signal 255. This may be done by generating binary values that indicate whether spectral magnitudes of the delayed audio signal 255 exceed thresholds. Measures of signal level 523 derived from the spectral magnitudes is passed to the external sound detector 510. These measures of signal level 523 may be the spectral magnitudes themselves.
  • the magnitudes of spectral components may be obtained by calculating the magnitudes of the transform coefficients generated by the filterbank 310b in the analyzer 300.
  • the magnitude of each transform coefficient is compared to a respective threshold that represents a frequency-dependent level at which a spectral component is deemed large enough to allow reliable calculations of system response characteristics. This process may be expressed as:
  • Vo(k) 1 indicates a significant transform coefficient k in delayed audio signal
  • the external sound detector 510 determines whether spectral components of ambient sounds have magnitudes that are too low to distort the estimated spectral magnitude response correction signal. This may be done by comparing transform coefficient magnitudes of the detection audio signal 235 with adjusted transform coefficient magnitudes of the second detection audio signal 275. The transform coefficient magnitudes of the second detection audio signal 275 are adjusted by an attenuation factor representing the degree of isolation between sounds in the acoustic channel through which the sound field 225 propagates and sounds outside this acoustic channel. The degree of isolation is estimated during design or setup of the system. The external sound detector 510 generates a set of values 515 that indicate whether the values of transform coefficients of the detected audio signal are essentiall immune to any ambient sounds that are present. This process may be expressed as:
  • V3 ⁇ 4fc) 1 indicates transform coefficient k of the detected audio signal is immune to ambient sounds
  • 3 ⁇ 4,iog(&) log magnitude of transform coefficient k in the detected audio signal
  • 3 ⁇ 4,iog(&) log magnitude of transform coefficient k in second detected audio signal
  • 3 ⁇ 4og(&) log magnitude of acoustic channel isolation for transform coefficient k.
  • the vector multiplier 530 multiplies the sets of values 515 and 525 and generates measures of validity 595 that indicate which components of the estimated spectral magnitude response correction signal 395 are deemed to be invalid or unreliable.
  • these measures comprise N/2+1 binary- valued elements where zero indicates a respective component is not reliable.
  • Measures of validity that are derived from the second detection audio signal 275 and the magnitudes of acoustic isolation as discussed above can be replaced or augmented by measures derived from an estimated acoustic error signal.
  • the estimated acoustic error signal can be obtained from a difference between corresponding spectral components of the detection audio signal 235 and the delayed audio signal 255.
  • the accuracy of the estimated acoustic error signal can be improved if the audio signal that is either passed into or received from the delay 250 is equalized by a minimum-phase filter having the target response.
  • This works well if the overall transfer function of the acoustic output transducer 220, the acoustic channel through which the sound field 225 propagates, and the acoustic input transducer 230 can be represented accurately by a minimum phase filter with a possible time delay.
  • the filtering effects of the AEQ filter 100 make the detection audio signal 235 substantially equal to the delayed audio signal 255 plus any ambient sounds that are detected by the acoustic input transducer 230.
  • the adaptation controller 410 in the control- signal generator 400 responds to the measures of validity 595 by doing either one or both of modifying components of the estimated spectral magnitude response correction signal 395 or adapting the leak factor vector 415.
  • the intention of these modifications is to eliminate or reduce the influence of ambient sounds and other noise on the adaptation of the AEQ filter 100. This may be done in a variety of ways. A few examples are described below.
  • the processes described below refer to bands that contain multiple components. These processes may also be used in implementations where the bands contain all the components for the estimated spectral magnitude response correction signal 395 or where a band contains only one component.
  • the following examples use the set of binary valued measures of validity 595 described above in which a value of one indicates a respective component of the estimated spectral magnitude response correction signal 395 is reliable and in which a value of zero indicates the component is not reliable.
  • the adaptation controller 410 inhibits adaptation of the AEQ filter 100 for a particular band if the measures of validity 595 indicate a majority of the components in the band are unreliable. This may be done in either of two ways.
  • the adaptation controller 410 continues passing the same control values for all components in that band until the measures of validity 595 indicate at least a majority of the components are reliable.
  • the adaptation controller 410 passes the new component values and the smoothing filter in the control- signal generator 400 generates a sequence of equalization-filter control signals 495 with control values for the band that smoothly change from the old values to the desired new values. This technique is sometimes referred to as a time-based zero-order hold.
  • the hold can be triggered in response to the measures of validity 595 indicating all components in a band are unreliable, or indicating one or more components in the band are unreliable.
  • the adaptation controller 410 sets the appropriate factors in the leak factor vector 415 to zero.
  • the adaptation controller 410 sets the appropriate factors to their customary non-zero values.
  • the adaptation controller 410 generates substitute values for unreliable components. This may be done in several ways. One way obtains substitute values by interpolating between values of reliable components. The interpolation is done across frequency and may be done using a first-order or linear interpolation between two reliable components or using a higher-order interpolation between a larger number of components. Another way obtains the substitute value from the nearest component that is reliable. This way is useful for components at band edges when interpolation is not possible.
  • the value of unreliable components may be modified to limit the variation between adjacent components in the frequency domain. This approach can be effective because errors can manifest themselves as significant localized deviations while any practical desired equalization has limited variations between adjacent components.
  • component-to-component variation can vary as a function of frequency or be constant. Suitable limits may be determined empirically by deriving system transfer functions 375 for a variety of listening environments and identifying the maximum component-to-component variation within all of the system transfer functions 375.
  • the second exemplary implementation is similar to the first implementation described above. The differences in implementation arise from what is analyzed in the analyzer 300. The second exemplary implementation is preferred for many applications because it is generally less sensitive to noise.
  • the analyzer 300 for this implementation performs substantially the same processes as those described above but, instead of comparing magnitudes of spectral components to derive the system transfer function 375, it compares averages of cross correlation and autocorrelation scores for those spectral components.
  • One implementation of the analyzer 300 is illustrated in Fig. 3. The following paragraphs describe differences with the
  • the log average cross correlation (LOG AVX) calculator 340a receives a block 315a of transform coefficients for the detection audio signal 235 from the filterbank 310a, receives a block 315b of transform coefficients for the delayed audio signal 255 from the filterbank 310b, calculates cross correlation scores for the transform coefficients in these two blocks, calculates an average for a series of the correlation scores using a leaky integrator, and obtains a logarithmic representation 355a of the averages.
  • conversion to a log domain is done to improve efficiency of some arithmetic calculations. Alternatively, these calculations as well as other calculations may be done in a linear domain.
  • the log average autocorrelation (LOG AVA) calculator 340b receives a block 315b of transform coefficients for the delayed audio signal 255 from the filterbank 310b, calculates autocorrelation scores for the transform coefficients in this block, calculates an average for a series of the autocorrelation scores using a leaky integrator, and obtains a logarithmic representation 355b of the averages.
  • LOG AVA log average autocorrelation
  • the complex conjugate component 341 obtains the complex conjugate of each transform coefficient in a block 315b of transform coefficients for the delayed audio signal 255 received from the filterbank 310b and passes this result to the vector multiplier 342.
  • the vector multiplier 342 also receives from the filterbank 310a a block 315a of transform coefficients for the detection audio signal 235 as its second input and calculates a block of cross correlation scores for its two inputs.
  • the block of correlation scores is passed to the vector multiplier 343, which attenuates each score in the block in response to respective factors in a first correlation leak factor vector 352.
  • the block of attenuated correlation scores is added to an attenuated block of average correlation scores received from the vector multiplier 348, and the resulting sum is passed to the delay 345, which imposes a one-sample delay on the block.
  • the delayed block of scores then becomes the new average of block correlation scores 346, which is passed to the vector multiplier 348 and to the log calculator 347.
  • the vector multiplier 348 attenuates each score in the block of average correlation scores in response to respective factors in a second correlation leak factor vector 353.
  • the log calculator 347 calculates the logarithmic representation 355a of each score in the block of average correlation scores.
  • the vector multiplier 343 and the vector multiplier 348 implement a conventional first-order low pass filters in which the factors in the two correlation leak factor vectors are related ows:
  • Convergence in adaptation may be improved by setting each f ⁇ (k) at least two times larger than the corresponding component of the leak factor vector 415. Smaller values of f ⁇ (k) result in adaptation overshoots and/or ringing.
  • xcorr (k.m) X A (k, m) - X D * (k, m) (7)
  • X D * (k, m) complex conjugate of transform coefficient k in block m of the delayed audio signal
  • the average of the cross correlation scores is calculated according to the following:
  • avexcorr(k,m) average cross correlation score for component k for blocks m.
  • the LOG AVA calculator 340b is also implemented as shown in Fig. 5 and operates the same as the LOG AVX calculator 340a except for the differences described here. Both of the LOG AVA calculator's inputs receive the same block 315b of transform coefficients for the delayed audio signal 255 from the filterbank 310b.
  • the vector multiplier 342 calculates a block of autocorrelation scores for the block of transform coefficients.
  • the average of the auto correlation scores is calculated according to the following:
  • the factors in the first and second correlation leak factor vectors may also be adapted.
  • the factors are adapted in response to values that represent the amount of noise that is present in the detection audio signal 235 such that the factor in the first correlation leak factor vector are smaller when more noise is present.
  • these factors may be adapted in response to the measures of validity 595 by setting the factor in the first vector for transform k to zero when the corresponding measure of validity 595 for the transform k indicates it is not reliable.
  • the factors in the second correlation leak factor vector are adapted according to expression 6.
  • the subtractor 370 subtracts the logarithmic values of the auto correlation scores received from the LOG AVA calculator 340b from the logarithmic values of the cross correlation scores received from the LOG AVX calculator 340a. This provides a
  • the subtracter 370 obtains a logarithmic representation of the system transfer function 375 by calculations that may expressed as:
  • the subtractor 390 subtracts the log values of the system transfer function 375 from the log values of the system target response 380 to obtain an estimated spectral magnitude response correction signal 395 as discussed above.
  • a further improvement in tradeoff between adaptation convergence and reduced sensitivity to noise can be obtained by varying the correlation leak factor vector 415 in response to the ratio of the levels of ambient sounds to the desired levels in the detection audio signal 235.
  • each component in the first correlation leak factor vector 352 is reduced from a nominal correlation leak factor value.
  • This ratio is calculated for each transform coefficient k in the log domain from the difference between the average log magnitude of transform coefficient k for the detection audio signal 235 and its expected log magnitude.
  • the expected log magnitude may be derived from the log magnitude of transform coefficient k for the delayed audio signal 255, the previously estimated system transfer function 375, and the system target response 380.
  • TGT log (k) log value of component k in the system target response
  • the log average autocorrelation scores for the component k of the delayed audio signal 255 may be obtained from the logarithmic representation 355b of averages generated by the LOG AVA calculator 340b.
  • the log average autocorrelation scores for the component k of the detection audio signal 235 may be obtained from an additional LOG AVA calculator that is applied to blocks 315a of transform coefficients for the detection audio signal 235 received from the filterbank 310a.
  • const is typically in the range from 0 to 10 dB and may be adjusted empirically to obtain a desired level of AEQ filter adaptation accuracy even with significant levels of ambient sound.
  • a larger value for const causes the averaging process to calculate averages over longer periods of time, which should reduce sensitivity to ambient sounds and increase adaptation accuracy.
  • the measures of validity 595 may be calculated using the detection audio signal 235, the delayed audio signal 255 and the second detection audio signal 275 as discussed above.
  • the value for the acoustic channel isolation 3 ⁇ 4 log (fc) for each transform coefficient k as shown in expression 5 will be larger in typical applications because the correlation method is less sensitive to noise.
  • the alternate approach mentioned above may also be used. Its use is generally more suitable in this second implementation because the use of correlation scores rather than spectral magnitudes decreases sensitivity to ambient sounds. As explained above, the use of correlation scores allows the alternate approach to provide good results even if the overall transfer function of the acoustic output transducer 220, the acoustic channel through which the sound field 225 propagates, and the acoustic input transducer 230 cannot be represented accurately by a minimum phase filter with a possible time delay.
  • the third exemplary implementation is similar to the first implementation described above. The differences in implementation arise from the fact the input to the delay 250 follows the AEQ filter 100 and, as a result, the response characteristics of this filter are not included in the system transfer function 375 that is derived in the analyzer 300. a) Signal Analysis
  • the analyzer 300 in the second implementation is substantially the same as the analyzer 300 in the first implementation; however, the estimated spectral magnitude response correction signal 395 in this implementation represents an estimate of the desired frequency response of the AEQ filter 100 instead of a correction to that response.
  • the subtractor 370 subtracts the logarithmic values of the transform coefficients received from the log calculator 320b from the logarithmic values of the transform
  • the system transfer function 375 that is calculated in this implementation does not include the response characteristics of the AEQ filter 100.
  • the resulting estimated correction signal 395 is an estimate of what the response characteristics of the AEQ filter 100 should be to obtain an overall system response that is equal to the system target response 380.
  • the estimated spectral magnitude response correction signal 395 is accurate over long time intervals but it can vary significantly from block to block.
  • the component values in the estimated correction signal 395 for any given block may differ significantly from the corrective values that should be provided to the AEQ filter 100.
  • a smoothing filter is applied to the estimated correction signal 395 to reduce or eliminate spurious variations. Smoothing is optional.
  • a smoothing filter is applied in the control-signal generator 400.
  • the control- signal generator 400 generates the equalization-filter control signal 495 that controls adaptation of the AEQ filter 100.
  • the control- signal generator 400 derives the equalization-filter control signal 495 by applying a smoothing filter to the estimated correction signal 395 to reduce system sensitivity to noise.
  • the control- signal generator 400 may also modify the values of selected
  • Fig. 7 illustrates one suitable implementation that includes a first-order smoothing filter in the form of a leaky integrator. Empirical tests have shown that a first order filter is adequate for many applications but higher order smoothing filters may be used if desired.
  • the implementation shown in Fig. 7 is similar to the implementation shown in Fig. 6 and discussed above.
  • the implementation used here includes an additional vector multiplier 460 that is controlled by a second leak factor vector 417 provided by the adaptation controller 410.
  • the vector multiplier 420 attenuates the magnitudes of the components in the current desired AEQ filter response 450 that are passed to the vector adder 430.
  • the vector multiplier 420 and the vector multiplier 460 implement a conventional first-order low pass filters in which the factors in the two correlation leak factor vectors are related to each other as shown above in expression 6.
  • the leak factor vector 415 may comprise factors with fixed values. If the leak factor vector 415 comprises factors with fixed values, these values may be chosen to provide a desired rate of adaptation of the AEQ filter 100 as described above.
  • the validity-measure generator 500 may be implemented as described above for the first implementation.
  • the fourth exemplary implementation shares features with the second and third implementations described above. The differences that are due to changes in the analysis performed in the analyzer 300 and in the control-signal generator 400 also apply to this fourth implementation.
  • the AEQ filter 100 may be essentially any type of filter structure including recursive, non-recursive and lattice structures provided it can adapt the magnitude of its frequency response according to the response characteristics specified by the equalization-filter control signal 495.
  • the AEQ filter 100 may be implemented by a bank of bandpass filters with overlapping or nearly overlapping passbands and respective gains for each bandpass filter.
  • the AEQ filter 100 may operate according to a set of filter parameters selected from multiple sets of predefined parameters in which each set provides a particular frequency response. The set of parameters that provides the closest match to the response specified by the equalization-filter control signal 495 is selected.
  • Another method derives appropriate filter parameters from the response characteristic specified by the equalization-filter control signal 495. Techniques that may be used to design sets of filter parameters for either method are discussed in international patent application publication no. WO 2010/014663 published Feb. 4, 2010.
  • This technique uses a block transform to implement the frequency-domain equivalent of convolving a block of signal samples with a finite impulse response. Essentially any time- domain to frequency-domain block transform and its inverse frequency-domain to time- domain block transform may be used.
  • the transform length is denoted by the symbol N.
  • the pad component 110 receives a segment 105 of N/2 samples for the audio signal 5 and appends it with a segment 107 of N/2 zero-valued samples to form a block 115 of N samples.
  • the FFT 120 is applied to the block 115 of samples to generate a block 125 of N transform coefficients.
  • the vector multiplier 134 receives the block 125 of transform coefficients and a block
  • the Inverse Fast Fourier Transform (IFFT) 144 is applied to the block 135 of filtered transform coefficients to generate a block of N time-domain signal samples.
  • the first half of this block of time-domain samples contains a segment 145 of N/2 samples that represent the initial response of the FFT 120 to the samples in the block 115.
  • the last half of the block of time- domain samples contains a segment 146 of N/2 samples that represent the ending response of the FFT 120 to the samples in the block 115.
  • the vector multiplier 131 receives a previous block 125 of transform coefficients through the delay 121 and a block 189 of values representing spectral magnitudes of a desired frequency response, and multiplies the magnitudes of the delayed transform coefficients with respective values in the desired frequency response to generate a block 132 of delayed filtered transform coefficients.
  • the delay 121 imposes a delay equal to the interval of one segment.
  • the IFFT 141 is applied to the block 132 of delayed filtered transform coefficients to generate a block of N time-domain signal samples.
  • the first half of this block of time- domain samples contains a segment 142 of N/2 samples that represent the initial response of the FFT 120 to the samples in the previous block 115.
  • the segment 142 is not used in this implementation.
  • the last half of the block of time-domain samples contains a segment 143 of N/2 samples that represent the ending response of the FFT 120 to the samples in the previous block 115.
  • the overlap-add component 151 adds the samples in the segment 145 to the samples in the segment 143 and outputs the resulting sum as a segment 152 of N/2 samples.
  • This segment is a portion of the filtered audio signal 195 when the frequency response of the AEQ filter 100 is static as is being described in these paragraphs.
  • the structures of some of the blocks and segments for this process are illustrated schematically in Figs. 9 and 10.
  • the samples in the segment 105 of N/2 audio signal samples are denoted by the symbol A(m), where m is a monotonically increasing block number. The block number increases by one for each subsequent block.
  • the samples in the segment 107 of N/2 zero-valued samples are denoted by the symbol Z(m).
  • the block of N time-domain signal samples generated by the IFFT comprises a segment of N/2 samples denoted by the symbol A(m,r) appended to a segment of N/2 samples denoted by the symbol Z(m,r), where r is an index for the filter frequency response.
  • the samples A(m,r) represent the FFT' s initial response to the samples in block m when the filter frequency response conforms to some response characteristic denoted by the index r.
  • the samples Z(m,r) represent the FFT' s ending response to the samples in block m when the filter frequency response conforms to the response characteristic r.
  • the overlap-add component 151 adds respective samples Z(m- l,r) and A(m,r) to obtain a segment 152 of N/2 interim samples denoted as IS(m,r).
  • the index r in this notation does not change if the response characteristics of the filter are static. If the response characteristics of the filter are to be adapted, the index r will increase by one for each requested change. The filter response may change as often as every block by using the additional processes described below.
  • Fig. 12 provides a schematic illustration of a sequence of processing steps that may be used to adapt the frequency response of the AEQ filter 100 when it is implemented as illustrated by the schematic block diagram in Fig. 11.
  • each segment 105 of the audio signal 5 is processed as discussed above to generate a respective block of time-domain signal samples.
  • Segments 153 and 157 are generated by processing the segment of samples A(m-l) with filter components adapted to conform to a desired frequency response represented by the index r-1.
  • Segments 148 and 149 are generated by processing the segment of samples A(m) with filter components adapted to conform to the same frequency response r-1.
  • Samples in the segment 157 are overlapped and added to samples in the segment 148 to obtain a segment 159 of interim samples IS(m,r-l).
  • a windowing operation 164 using the last half of a window function WF is applied to the interim samples IS(m,r- l) in the segment 159. This last half of the window function is represented by the symbol WF2.
  • a suitable window function WF2 is described below.
  • Segments 142 and 143 are generated by processing the segment of samples A(m-l) with filter components adapted to conform to a desired frequency response represented by the index r.
  • Segments 145 and 146 are generated by processing the segment of samples A(m) with filter components adapted to conform to the same frequency response r.
  • Samples in the segment 143 are overlapped and added to samples in the segment 145 to obtain a segment 155 of interim samples IS(m,r).
  • a windowing operation 161 using the first half of the window function WF is applied to the interim samples IS(m,r) in the segment 155. This first half of the window function is represented by the symbol WFl .
  • the samples in the two window- weighed segments are overlapped and added to obtain a segment 169 of N/2 samples in the filtered audio signal 195.
  • the half- window functions WFl and WF2 add to one when overlapped with one another, and provide good frequency selectivity and stop-band rejection.
  • Suitable window functions WFl and WF2 are defined by the following expressions:
  • Fig. 11 is a schematic block diagram of a device that may be used to perform the process just described. This diagram is not intended to illustrate a practical implementation. For example, some of the FFT and IFFT components can be eliminated by using buffers to store time-domain samples calculated for previous segments of samples.
  • the vector multiplier 136 receives the block 125 of transform coefficients and a block 189 of values delayed by an interval of one segment representing spectral magnitudes of a previous desired frequency response, and multiplies the magnitudes of the transform coefficients with respective values in the previous desired frequency response to generate a block 137 of filtered transform coefficients.
  • the IFFT 147 is applied to the block 137 of filtered transform coefficients to generate a block of N time-domain signal samples.
  • the first half of this block of time-domain samples contains a segment 148 of N/2 samples that represent the initial response of the FFT 120 to the samples in the block 115.
  • the last half of the block of time-domain samples contains a segment 149 of N/2 samples that represent the ending response of the FFT 120 to the samples in the block 115.
  • the segment 149 is not used in this implementation.
  • the overlap-add component 154 adds the samples in the segment 145 to the samples in the segment 143 and outputs the resulting sum as a segment 155 of N/2 samples.
  • a windowing operation 161 is applied to the samples in the segment 155 using the first half of the windowing function WF to generate a segment 162 of N/2 windowed samples.
  • the first half of this function is represented by the symbol WFl.
  • the delay 156 imposes a delay of one segment interval to the segment 146 to generate a segment 157 of N/2 delayed samples.
  • the overlap-add component 158 adds the samples in the segment 148 to the delayed samples in the segment 157 and outputs the resulting sum as a segment 159 of N/2 samples.
  • a windowing operation 164 is applied to the samples in the segment 159 using the last half of the windowing function WF to generate a segment 165 of N/2 windowed samples. The last half of this function is represented by the symbol WF2.
  • the vector adder 167 overlaps and adds corresponding samples in the segments 162 and 165 to generate a segment 169 of N/2 samples in the filtered audio signal 195.
  • the remaining components 181 to 186 shown in the drawing are applied to the equalization-filter control signal 495 to generate the blocks 189 of values representing spectral magnitudes of desired frequency responses.
  • the antilog 181 generates linear-domain magnitudes of the components in the equalization-filter control signal 495.
  • the Hilbert transform 182 is applied to the log magnitude values of the equalization-filter control signal 495 to generate a set of angular coefficients.
  • the vector multiplier 183 multiplies each components of the linear-domain magnitudes with a respective angular coefficient to generate a set of complex- valued coefficients in a frequency-domain representation of a minimum-phase causal impulse response that has the same frequency response as that specified in the equalization-filter control signal 495. Additional details of this process may be obtained from Oppenhiem et al., Digital Signal Processing, Prentice Hall Inc., 1975, pp 337-361
  • the IFFT 184 is applied to the set of coefficients in the frequency-domain
  • any smoothly-varying window function with a length no greater than N/2+1 may be used. Window functions with abrupt variations are not desirable because their use will introduce audible artifacts into the derived frequency response.
  • window function WX may be used:
  • Zero-valued samples are appended to the windowed impulse response to obtain a block of N samples.
  • the FFT 186 is applied to the windowed and appended impulse response to generate a block 189 of values representing spectral magnitudes of the desired frequency response that is specified by the equalization-filter control signal 495.
  • the filtering operations that are performed in the AEQ filter 100 and the analyzer 300 may be implemented by essentially any filtering technology that may be desired.
  • the four exemplary implementations discussed above use the FFT and IFFT computational methods to implement DFT and IDFT filterbanks.
  • An alternative implementation described here uses the FFT and IFFT computational methods to implement MDFT and IDFT filterbanks.
  • the MDFT filterbank generates blocks of transform coefficients according to the following expression:
  • the first Nil transform coefficients X(k) in each block are unique and have complex values.
  • the complementary IMDFT filterbank generates blocks of time-domain samples according to the following expression:
  • a pair of windowing operations are used with these filterbanks.
  • One windowing operation applies an analysis window function to the audio signal prior to an analysis or forward transform.
  • Another windowing operation applies a synthesis window function to blocks of time-domain samples generated by a synthesis or inverse transform.
  • any window functions may be used but the analysis and synthesis window functions should be designed so that their product window, when overlapped with itself by half its length, adds to one.
  • One exemplary function that may be used for each of the analysis and window functions is the sine function with its domain scaled so that zero to pi radians corresponds to 0 to N-l samples.
  • a windowing operation 610 applies the window function WQ to a block of N samples of the audio signal 5 to generate a block 615 of N window- weighted samples.
  • the FFT 620 is applied to the block 615 of window- weighted samples to generate a block 625 of N transform coefficients.
  • the vector multiplier 630 receives the block 625 of transform coefficients and a block 685 of values representing spectral magnitudes of a desired frequency response, and multiplies the magnitudes of the transform coefficients with respective values in the desired frequency response to generate a block 635 of filtered transform coefficients.
  • the Inverse Fast Fourier Transform (IFFT) 640 is applied to the block 635 of filtered transform
  • a sequence of the blocks 645 of time-domain samples are each weighted by the window function WQ, overlapped with one another by N/2 samples, and corresponding samples in overlapped blocks are added.
  • This windowing-overlap-add process may be performed in a variety of ways. One way is illustrated in the figure.
  • the delay 650 imposes a delay of N/2 samples on a block 645 of time-domain samples to generate a delayed block 655 of time-domain samples.
  • the windowing operation 661 applies the first half WQl of the window function WQ to the first half of the block 645 of time-domain samples to generate a segment 664 of N/2 windowed time-domain samples.
  • the windowing operation 662 applies the last half WQ2 of the window function WQ to the last half of the delayed block 655 of time-domain samples to generate a delayed segment 665 of N/2 windowed time-domain samples.
  • the overlap-add component 670 adds corresponding samples in the segment 664 of windowed time-domain samples and the delayed segment 665 of windowed time-domain samples and outputs the resulting sums as a segment of N/2 time-domain samples in the filtered audio signal 195.
  • the antilog 680 is applied to the components in the equalization-filter control signal 495 to generate a block 189 of values in the linear-domain that represent spectral magnitudes of the desired frequency response that is specified by the equalization-filter control signal 495.
  • An alternative implementation of the analyzer 300 omits the filterbank 310b and receives a frequency-domain representation of the delayed audio signal directly from the delay 250. This may be achieved by obtaining a frequency-domain representation of the audio signal 5 from the AEQ filter 100 and passing this representation to the delay 250.
  • the amount of delay that can be imposed by the delay 250 in this implementation is equal to the interval of an integer number of segments, where each segment has Nil samples. If desired, greater accuracy in the temporal alignment between the delayed audio signal 255 and the detection audio signal 235 may be achieved by ensuring the delay imposed by the delay 250 is greater than what is required to achieve proper alignment and introducing an additional delay somewhere in the signal path between the acoustic input transducer 230 and the filterbank 310a. The additional delay can be implemented using either analog or digital techniques to obtain the desired temporal alignment.
  • the spectral resolution of the AEQ filter 100 should be high enough to provide good equalization for the most demanding system transfer function it is likely to encounter. This spectral resolution is determined by the length N of the FFT 620 and the shape of the analysis window function used in the windowing operation 610. For a given analysis window function, spectral resolution increases as the transform length increases.
  • Signal processing delays in the AEQ filter 100 also increase as the transform length increases.
  • the technique discussed below provides a way to decrease processing delays for a given transform length.
  • the window function WQ is set to 180 samples.
  • the windowing operation 610 generates sample blocks 615 having a length equal to 256 that each comprise 76 zero-valued samples appended to 180 window- weighted samples of the audio signal 5.
  • the FFT 620 and the IFFT 640 each have a length equal to 256.
  • the IFFT 640 generates blocks 645 each having 256 time-domain samples. The last 76 samples in each block can be ignored.
  • the windowing operation 661 is applied to the first half of the first 180 samples in each block and the window operation 662 is applied to the last half of the first 180 samples in the block.
  • the overlap-add component 670 adds corresponding samples in the segment 664 of windowed time-domain samples and the delayed segment 665 of windowed time-domain samples and outputs the resulting sums as a segment of 90 time-domain samples in the filtered audio signal 195.
  • FIG. 14 is a schematic block diagram of a device 70 that may be used to implement aspects of the present invention.
  • the processor (PROC) 72 provides computing resources.
  • RAM 73 is system random access memory (RAM) used by the PROC 72 for processing.
  • ROM 74 represents some form of persistent storage such as read only memory (ROM) for storing programs needed to operate the device 70 and possibly for carrying out various aspects of the present invention.
  • I/O control 75 represents interface circuitry to receive and transmit signals such as the signals 5, 195, 235, 255, 275 and 495 mentioned above.
  • bus 71 which may represent more than one physical or logical bus; however, a bus architecture is not required to implement the present invention.
  • additional components may be included for interfacing to devices such as a keyboard or mouse and a display, and for controlling a storage device 78 having a storage medium such as magnetic tape or disk, or an optical medium.
  • the storage medium may be used to record programs of instructions for operating systems, utilities and applications, and may include programs that implement various aspects of the present invention.
  • Software implementations of the present invention may be conveyed by a variety of machine readable media such as baseband or modulated communication paths throughout the spectrum including from supersonic to ultraviolet frequencies, or storage media that convey information using essentially any recording technology including magnetic tape, cards or disk, optical cards or disc, and detectable markings on media including paper.
  • machine readable media such as baseband or modulated communication paths throughout the spectrum including from supersonic to ultraviolet frequencies, or storage media that convey information using essentially any recording technology including magnetic tape, cards or disk, optical cards or disc, and detectable markings on media including paper.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Acoustics & Sound (AREA)
  • Signal Processing (AREA)
  • Soundproofing, Sound Blocking, And Sound Damping (AREA)

Abstract

L'invention porte sur des techniques à domaine de fréquence qui sont utilisées pour une égalisation adaptative qui est sensible à des caractéristiques d'amplitude spectrale mais non sensible à des caractéristiques de phase d'une réponse de système. Une corrélation de signal peut être utilisée pour améliorer une précision d'adaptation lorsque des niveaux significatifs de sons ambiants sont présents. Une mise en œuvre de filtre préférée utilise des transformées par bloc à base de convolution et des fenêtres de fondu enchaîné.
PCT/US2011/051322 2010-10-14 2011-09-13 Égalisation automatique à l'aide d'un filtrage adaptatif par domaine de fréquence et d'une convolution rapide dynamique WO2012050705A1 (fr)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN201180049284.8A CN103155591B (zh) 2010-10-14 2011-09-13 使用自适应频域滤波和动态快速卷积的自动均衡方法及装置
EP11764382.5A EP2628317B1 (fr) 2010-10-14 2011-09-13 Egalisation automatique avec filtrage adaptatif dans le domaine fréquentiel et convolution rapide dynamique
US13/878,705 US9084049B2 (en) 2010-10-14 2011-09-13 Automatic equalization using adaptive frequency-domain filtering and dynamic fast convolution

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US39322410P 2010-10-14 2010-10-14
US61/393,224 2010-10-14

Publications (1)

Publication Number Publication Date
WO2012050705A1 true WO2012050705A1 (fr) 2012-04-19

Family

ID=44736043

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2011/051322 WO2012050705A1 (fr) 2010-10-14 2011-09-13 Égalisation automatique à l'aide d'un filtrage adaptatif par domaine de fréquence et d'une convolution rapide dynamique

Country Status (4)

Country Link
US (1) US9084049B2 (fr)
EP (1) EP2628317B1 (fr)
CN (1) CN103155591B (fr)
WO (1) WO2012050705A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2014186106A1 (fr) * 2013-05-16 2014-11-20 Apple Inc. Egalisation audio adaptative pour dispositifs d'écoute personnels
US10142763B2 (en) 2013-11-27 2018-11-27 Dolby Laboratories Licensing Corporation Audio signal processing

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2969804A1 (fr) * 2010-12-23 2012-06-29 France Telecom Filtrage perfectionne dans le domaine transforme.
US9466305B2 (en) 2013-05-29 2016-10-11 Qualcomm Incorporated Performing positional analysis to code spherical harmonic coefficients
US9502044B2 (en) 2013-05-29 2016-11-22 Qualcomm Incorporated Compression of decomposed representations of a sound field
US9922656B2 (en) 2014-01-30 2018-03-20 Qualcomm Incorporated Transitioning of ambient higher-order ambisonic coefficients
US9502045B2 (en) 2014-01-30 2016-11-22 Qualcomm Incorporated Coding independent frames of ambient higher-order ambisonic coefficients
US9852737B2 (en) 2014-05-16 2017-12-26 Qualcomm Incorporated Coding vectors decomposed from higher-order ambisonics audio signals
US9620137B2 (en) 2014-05-16 2017-04-11 Qualcomm Incorporated Determining between scalar and vector quantization in higher order ambisonic coefficients
US10770087B2 (en) 2014-05-16 2020-09-08 Qualcomm Incorporated Selecting codebooks for coding vectors decomposed from higher-order ambisonic audio signals
CN103987001A (zh) * 2014-05-28 2014-08-13 深圳市金立通信设备有限公司 一种音频修正的方法及装置
CN103987000A (zh) * 2014-05-28 2014-08-13 深圳市金立通信设备有限公司 一种音频修正的方法及终端
US9747910B2 (en) 2014-09-26 2017-08-29 Qualcomm Incorporated Switching between predictive and non-predictive quantization techniques in a higher order ambisonics (HOA) framework
CN107493247B (zh) * 2016-06-13 2021-10-22 中兴通讯股份有限公司 一种自适应均衡方法、装置及均衡器
US10644820B2 (en) * 2017-02-06 2020-05-05 Huawei Technologies Co., Ltd. Waveform-coding for multicarrier wake up radio frame
CN110161773B (zh) * 2019-04-26 2021-11-02 太原理工大学 基于切割超连续谱的超宽带白噪源
TW202105908A (zh) * 2019-06-26 2021-02-01 美商杜拜研究特許公司 具有改善頻率解析度的低延遲音訊濾波器組
US11489505B2 (en) * 2020-08-10 2022-11-01 Cirrus Logic, Inc. Methods and systems for equalization

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4207624A (en) * 1978-10-02 1980-06-10 Rockwell International Corporation Frequency domain adaptive filter for detection of sonar signals
US5222189A (en) 1989-01-27 1993-06-22 Dolby Laboratories Licensing Corporation Low time-delay transform coder, decoder, and encoder/decoder for high-quality audio
US20090220105A1 (en) * 2005-05-01 2009-09-03 Harry Bachmann Method for compensating for changes in reproduced audio signals and a corresponding device
WO2010014663A2 (fr) 2008-07-29 2010-02-04 Dolby Laboratories Licensing Corporation Procédé de contrôle adaptatif et égalisation de canaux électroacoustiques

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4939685A (en) 1986-06-05 1990-07-03 Hughes Aircraft Company Normalized frequency domain LMS adaptive filter
NL8601604A (nl) 1986-06-20 1988-01-18 Philips Nv Frequentie-domein blok-adaptief digitaal filter.
US6760451B1 (en) 1993-08-03 2004-07-06 Peter Graham Craven Compensating filters
JP2004507922A (ja) 2000-08-21 2004-03-11 コーニンクレッカ　フィリップス　エレクトロニクス　エヌ　ヴィ 区分ブロック周波数領域適応フィルタ
EP1314247B1 (fr) 2000-08-21 2007-01-17 Koninklijke Philips Electronics N.V. Filtre adaptatif domaine frequentiel blocs cloisonnes
US6868162B1 (en) 2000-11-17 2005-03-15 Mackie Designs Inc. Method and apparatus for automatic volume control in an audio system
FR2827730A1 (fr) 2001-07-17 2003-01-24 Koninkl Philips Electronics Nv Recepteur, procede, programme et signal de transport pour adapter le volume sonore d'un signal acoustique d'appel entrant
US7480377B2 (en) 2003-12-31 2009-01-20 Intel Corporation Dual adaptive filter apparatus and method
CA2471674A1 (fr) 2004-06-21 2005-12-21 Soft Db Inc. Systeme et methode de masquage sonore a reglage automatique
WO2011159858A1 (fr) 2010-06-17 2011-12-22 Dolby Laboratories Licensing Corporation Procédé et appareil pour réduire l'effet du bruit ambiant sur des auditeurs

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4207624A (en) * 1978-10-02 1980-06-10 Rockwell International Corporation Frequency domain adaptive filter for detection of sonar signals
US5222189A (en) 1989-01-27 1993-06-22 Dolby Laboratories Licensing Corporation Low time-delay transform coder, decoder, and encoder/decoder for high-quality audio
US20090220105A1 (en) * 2005-05-01 2009-09-03 Harry Bachmann Method for compensating for changes in reproduced audio signals and a corresponding device
WO2010014663A2 (fr) 2008-07-29 2010-02-04 Dolby Laboratories Licensing Corporation Procédé de contrôle adaptatif et égalisation de canaux électroacoustiques

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
CLARK G A ET AL: "A Unified Approach to Time- and Frequency-Domain Realization of FIR Adaptive Digital Filters", IEEE TRANSACTIONS ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING, IEEE INC. NEW YORK, USA, vol. ASSP-31, no. 5, 1 October 1983 (1983-10-01), pages 1073 - 1083, XP002123672, ISSN: 0096-3518 *
DENTINO,MCCOOL,WIDROW: "Adaptive Filtering in the Frequency Domain", PROCEEDINGS OF THE IEEE, vol. 66, no. 12, 1 December 1978 (1978-12-01), pages 1658 - 1659, XP002666092 *
OPPENHIEM ET AL.: "Digital Signal Processing", 1975, PRENTICE HALL INC., pages: 113 - 115
OPPENHIEM ET AL.: "Digital Signal Processing", 1975, PRENTICE HALL INC., pages: 337 - 361
SONG LIU ET AL: "Transform domain adaptive filter in active noise control", SIGNAL PROCESSING, 2002 6TH INTERNATIONAL CONFERENCE ON AUG. 26-30, 2002, PISCATAWAY, NJ, USA,IEEE, vol. 1, 26 August 2002 (2002-08-26), pages 272 - 275, XP010627977, ISBN: 978-0-7803-7488-1 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2014186106A1 (fr) * 2013-05-16 2014-11-20 Apple Inc. Egalisation audio adaptative pour dispositifs d'écoute personnels
US9515629B2 (en) 2013-05-16 2016-12-06 Apple Inc. Adaptive audio equalization for personal listening devices
US10142763B2 (en) 2013-11-27 2018-11-27 Dolby Laboratories Licensing Corporation Audio signal processing

Also Published As

Publication number Publication date
US9084049B2 (en) 2015-07-14
CN103155591B (zh) 2015-09-09
EP2628317B1 (fr) 2015-10-07
CN103155591A (zh) 2013-06-12
EP2628317A1 (fr) 2013-08-21
US20130208917A1 (en) 2013-08-15

Similar Documents

Publication Publication Date Title
US9084049B2 (en) Automatic equalization using adaptive frequency-domain filtering and dynamic fast convolution
US10354634B2 (en) Method and system for denoise and dereverberation in multimedia systems
JP5362894B2 (ja) 音声変換器の線形及び非線形歪みを補償するためのニューラル・ネットワーク・フィルタリング技術
US7602925B2 (en) Audio feedback processing system
EP2831871B1 (fr) Appareil et procédé permettant d'améliorer la qualité de reproduction sonore perçue en combinant l'annulation active de bruit et la compensation perceptuelle de bruit
EP2392149B1 (fr) Procédé de détermination d'un filtre inverse pour un haut-parleur
JP2010537586A (ja) 自動センサ信号整合
WO2009042385A1 (fr) Procédé et appareil pour générer un signal audio à partir de multiples microphones
US11580966B2 (en) Pre-processing for automatic speech recognition
CN103380628A (zh) 音频处理装置、音频处理方法和程序
Cecchi et al. An adaptive multiple position room response equalizer
JP4478045B2 (ja) エコー消去装置、エコー消去方法、エコー消去プログラムおよびその記録媒体
JPH0248831A (ja) 雑音低減方法
JP2012100117A (ja) 音響処理装置及び方法
Axelson-Fisk Caring More About EQ Than IQ: Automatic Equalizing of Audio Signals
JP2007174632A (ja) 伝達系推定装置、方法、プログラム、記録媒体
JP2007306294A (ja) 信号生成方法、信号生成装置およびインパルス応答測定方法

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 201180049284.8

Country of ref document: CN

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11764382

Country of ref document: EP

Kind code of ref document: A1

DPE2 Request for preliminary examination filed before expiration of 19th month from priority date (pct application filed from 20040101)
WWE Wipo information: entry into national phase

Ref document number: 2011764382

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 13878705

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE